Effect of Time Delay on the Transmission of Dengue Fever

World Academy of Science, Engineering Technology International Journal of Mathematical Computational Sciences Effect of Time Delay on the Transmission of Dengue Fever K. Patanarapelert I.M. Tang International Science Index, Mathematical Computational Sciences waset.org/publication/12037 Abstract The effect of a time delay on the transmission on dengue fever is studied. The time delay is due to the presence of an incubation period for the dengue virus to develop in the mosquito before the mosquito becomes infectious. The conditions for the existence of a Hopf bifurcation to limit cycle behavior are established. The conditions are different from the usual one they are based on whether a particular third degree polynomial has positive real roots. A theorem for determining whether for a given set of parameter values, a critical delay time exist is given. It is found that for a set of realistic values of the parameters in the model, a Hopf bifurcation can not occur. For a set of unrealistic values of some of the parameters, it is shown that a Hopf bifurcation can occur. Numerical solutions using this last set show the trajectory of two of the variables making a transition from a spiraling orbit to a limit cycle orbit. Keywords Dengue fever transmission, time delay, Hopf bifurcation, limit cycle behavior I I. INTRODUCTION. NSIGHTS into the behavior of systems can often be achieved through a mathematical modeling of the system. The models are usually expressed as a set of differential equations obtained from physical laws, Newton s second law or from the basic principle that the time rate of change of the number of members in a category is equal to the numbers entering minus the numbers leaving. The last method is the main one used to obtain models for chemical biochemical reactions. Mathematical modeling has undergone a renaissance in recent years. This has occurred because much of biology medicine is now discussed in terms of molecular biology processes. The full force of mathematical modeling can now be applied to these processes. In this paper, we are interested in the effects of a time delay caused by an incubation period in the virus development on the transmission of dengue fever. Dengue fever (DF) is an illness that is characterized by a moderately high fever, extreme pain in stiffness in the joints, a rash a reduction in the white blood cells [1]. These symptoms are Manuscript received January 23, 2008. Kot Patanarapelert was with the Department of Mathematics, Faculty of Science, Mahidol University. He is now with the 1 Department of Mathematics, Faculty of Science, Silpakorn University, Sunan Chra Palace Campus, Nakhon Pathum 73000, Thail I-Ming Tang is with the Department of Physics, Faculty of Science, Mahidol University, Bangkok 10400, Thail. (Corresponding author) Tel: 662 201 5758, email: scimt@mahidol.ac.th. caused by the toxins produced by one of the four serotypes of a virus belonging to the genus Flavirus, in the family Flavicidae. In many cases, the illness is asymptomatic an infection can only be determined through serologic tests. A second infection by this virus can result in a more virulent form of the disease, dengue hemorrhagic fever (DHF). From its first appearance in the Philippines in 1953, DHF has become the most important arthropod-borne viral diseases of human s [2]. It has been estimated that there are between 50 100 million cases of dengue fever (DF) a year, over 250,000 cases of dengue haemorrhagic fever (DHF) with approximately 10,000 infant deaths due to the latter form of this disease. Time delays can play a very import role in the dynamics of real systems. Martin Ruan [3] showed that a time delay in a generalized Gause-type predator-prey model could cause a stable equilibrium to become unstable. Khan Greenhalgh [4] have studied the consequences of there being a time delay in the effects of vaccination on the spread of diseases in an epidemic. Xiao Chen [5] have studied a predator-prey model showed that a time delay can cause in their model, a transition from a stable equilibrium to an unstable one then a transition back to a stable one. Other recent studies have been done by Ruan Wei [6] by Tam [7]. Most of the above studies have used the Hopf bifurcation method [8] to analysis their models. They did not, however, include all the effects of the time delay. By neglecting one term, they obtained equilibrium states, which do not depend on the time delay. In this paper, we included all the effects of the time delay find that the equilibrium state does depend on the time delay,. In Section II, we introduce a model for the transmission of the disease, which takes into account a delay in the mosquito becoming infectious after it is infected with the virus. This delay leads to the factor exp {- v } ( v being the death rate or inverse mean life of the mosquitoes) appearing in the equations describing the evolution of the mosquito populations. Since the mean life the delay (incubation period) time are of the same magnitude, the product v does not approach zero so the exponential factor can not be dropped, i.e. set to unity. We also obtain here the equilibrium states of the system the conditions for the endemic state to be physically realizable. In Section III, we establish the conditions under which a Hopf bifurcation to a limit cycle behavior is possible. We perform a parameter space analysis in Section IV, show that the conditions can not be met when realistic values of the International Scholarly Scientific Research & Innovation 1(10) 2007 490

World Academy of Science, Engineering Technology International Journal of Mathematical Computational Sciences International Science Index, Mathematical Computational Sciences waset.org/publication/12037 (biological) parameters are used. Using unrealistic values of the parameters however, we find that the conditions for bifurcation can be met. Solving the equations for these parameter values, we obtain trajectories in phase space which shows a stable spiral node for values of < c a limit cycle when = c ( c being the critical delay at which the equilibrium state becomes unstable bifurcates). Finally in Section V, we discuss some implications of our results. II. MATHEMATICAL MODEL OF DENGUE FEVER TRANSMISSION II.a Mathematical Model. To formulate a model for dengue fever transmission, one needs to know what the transmission cycle of this disease is. The infection in the human begins when an infectious mosquito bites a human injects a large number of dengue viruses of one strain into the blood of the human. There, the virus develops causes either a symptomatic or asymptomatic infection in the person. The illness resulting from the infection last for about one to two weeks. During this time, the infected person is immune to further infection by all of the four dengue virus strains. After the person recovers, he keeps his immunity to the infecting strain but losses the temporary immunity he had to the other strains. To simply matters, we have assumed in our model there is only one strain present. If a susceptible mosquito bites a person while he has a high count of virus in his blood, virus could enter into the mosquito mosquito is said to be infected. It then takes from 3 to 14 days (the incubation period) for the virus to develop inside the mosquito before the mosquito is able to transmit the disease to a human by its bite. To represent the transmission process, we divide the human population into three classes, susceptible, infectious recovered (S, I R ) the mosquito population into two classes, susceptible infectious, S v I v. The time rate of change in the number of subjects in each class is equal to the number of subjects entering into the group minus the number leaving the group. This gives for the different human population classes ds'(t) bh N S'(t)I T v( t) hs'(t) N c di'(t) bh S'(t)I' v( t) (h r)i'(t) NT c (1a) (1b) dr'(t) ri'(t) hr'(t). (1c) The time rate of change of the number of susceptible mosquitoes S v is ds' (t) b v v A S' (t) I'(t) S' (t) v v v N m T of the infected mosquitoes is h bhm (1d) di' b v(t) v -v S' v(t-) I'(t-)e vi' v(t) T N c (1e) Since we are interested in the time rate of change of the infectious mosquitoes at time t since it takes number of days for the infected mosquitoes to become infectious, we should be interested in the number of susceptible mosquitoes who bit an infected human at the time t -. not at the time t. Between the times t t -, a portion of these infected mosquitoes would have died. Taking into account all of these additional considerations, we get In the above, N T is the total host population; A, the recruitment rate of female mosquitoes;, the human birth rate; h ( v ), the death rate of the humans (mosquitoes); h ( v ), the probability that a bite by an infected mosquito (human) on a susceptible human (mosquito) will result in a new infection; r, the rate at which the infected human recovers; b, the biting rate of the mosquito c is the number of other animals the mosquitoes can fed on. The derivation of the contact term is given in ref. 9. Adding Eqns. (1a) - (1c) together Eqns. (1d) to (1e), we obtain dnt ( h )N T where NT S' I' R' (2a) dn v =A- v N v where Nv =S v '+I v ' (2b) If we assume that the total human mosquito population remains constant, we have h N V = A/ v. Dividing the human classes by the total human population the mosquito classes by the total mosquito populations, we get the densities for each class. We also have S + I + R = 1 S v + I v = 1 where the absence of the prime denotes a density. Because of these two constraints, only three equations are needed to define the model which we take to be ds di h hsiv hs, (3a) SI ( r)i (3b) h v h di v (t) -v vs v(t-)i(t)e vi v(t) (3c) where v bv (4a) h bhm (4b) with m being the ratio between the total number of mosquitoes total number of humans where we have set c = 0. In Eqn. (3c), we have replaced I(t-) by I(t) since the density of infectious humans is not expected to vary much over the period which is much less then the life time of a human. International Scholarly Scientific Research & Innovation 1(10) 2007 491

World Academy of Science, Engineering Technology International Journal of Mathematical Computational Sciences International Science Index, Mathematical Computational Sciences waset.org/publication/12037 II.b Equilibrium States Their Stabilities. The equilibrium states are obtained by setting the RHS of Eqns. (3a) to (3c) to zero. Doing this, we get two equilibrium states, the disease free state, E o =(0, 1, 0) the endemic equilibrium state, E 1 = (I v *, S*, I*) where v I* v v, I v* e (5a) v v I* e 1 where with v v I* e v 1 v S* v h r 1 Ro I*e v v h I* R v oe 1 v h r v R o e v h (5b) (5c) 2 b m v h Ro (5d) v(h r) As we see, S*, I* I v * are functions of the delay time. This would not be true if we had dropped the exponential factor, exp{- v }. Since I* 0, we need R o exp{- v } 1 (6) For the equilibrium point E 1 to exist, must lie in the range 0 (ln R o )/ v. The factor R = R o exp{- v } is called the basic reproduction number it is the number of secondary infections resulting from a primary infection. When it is less than one, the disease-free state is the equilibrium state; if it is greater one, then the endemic state is the equilibrium state. III, Bifurcation Conditions for the Endemic Equilibrium State. III.a Basic Theorems. To establish the conditions for the stability of the endemic state, we introduce the following two theorems: Theorem 1. The equilibrium point of Eqns. (3a) (3c) is stable if only if every characteristic root of the matrix J has a real part not greater than zero, those with zero real part are zero. It is asymptotically stable if only if every characteristic root of J has nega-tive real part. Theorem 2. Let x* = x*(a) be an equilibrium point of a system of first order differential equations given by dx F({x};a) where F( ) is a column vector let the Jacobian matrix at an equilibrium point be defined as F J( a) D F(x*,a) i x (x*,a) i,j 1,2,.,n, x j if J(a) has a pair of complex eigenvalues, (a) = u(a) iv(a) such that i. u(a c ) = 0, ii. v(a c ) = v* > 0 du iii. (a c ) 0 da where a c is called a critical value of the bifurcaion parameter a, no other eigenvalues with zero real part exist, the system undergoes a transition to a limit cycle about the point (x*,a c ). The proofs of these two theorems can be found in ref. [8]. III.b. Application to the Present System. Whether the equilibrium point of the given system undergoes a Hopf bifurcation to a limit cycle behavior is determined by whether the eigenvalues of the Jacobian for the system of equations satisfy the two theorems. The presence of a -dependence in the equations requires modifications of the usual conditions for a Hopf bifurcation. To see this, we diagonalizing the Jacobian obtain the following characteristic equation - P(, ) Q(, )e v 0 (8) where P() = 3 + a o () 2 + b o () + a 2 () (9a) Q() = a 1 () 2 + b 1 () - a 3 () (9b) with a o () = 2 h + r + h I v *, (10a) (10c) a 1 () = v I*exp{- v }, a 2 ()= ( h + h I v *)( v ( h r) v h S*exp{- v }) + v h 2 S*I v * exp{- v }, (10b) a 3 () = - exp{- v }(( h + h I v *) h r) v I* + h h v S*I v *), (10d) b o () = ( h + h I v *)exp{- v }( h v r) + v ( h r) v h S*exp{- v }, (10e) b 1 () = ((2 h + r + h I v *) v I* + v h S*I v *)exp{- v }. (10f) The characteristic equations obtained by Ruan Wei [6] by Klan Greenhalgh [4] for their models are of form 3 + a 2 + b + c = de -, (11a) while the characteristic equation studied by Tam [7] has the form 3 + a 2 + (b + ce - ) + d = fe - (11b) International Scholarly Scientific Research & Innovation 1(10) 2007 492

World Academy of Science, Engineering Technology International Journal of Mathematical Computational Sciences International Science Index, Mathematical Computational Sciences waset.org/publication/12037 The constants (a, b, c, d, f) in Eqns. (9) (10) are defined in the respective references. The important thing to note is that none of the constants depend on. To determine the conditions for Hopf bifurcation, we apply the technique used in refs. [4] [6]. Substituting = u + iv (where u v are real numbers may be functions of ) into Eqn. (7) separating the real imaginary parts, we get 3u()v() 2 - u() 3 - a o ()(u() 2 - v() 2 ) - b o ()u() - a 2 () = [{a 1 () (u() 2 - v() 2 ) + b 1 ()u() - a 3 ()}cos(v()) + {2a 1 ()u()v() + b 1 ()v()}sin(v())]e -u (12) v() 3-3u() 2 v()-2a o ()u()v() - b o ()v() = [{2a 1 ()u()v()+) + b 1 ()v()}cos(v()) {a 1 () (u() 2 - v() 2 ) + b 1 ()u() - a 3 ()}sin(v())]e -u. (13) We now set c. At this value of, u( c ) = 0. Denoting v( c ) as v*, Eqns (12) (13) become a o ( c )v* 2 - a 2 ( c ) = b 1 ( c )v*sin(v* c ) - (a 1 ( c )v* 2 + a 3 ( c ))cos(v* c ) v* 3 - b o ( c )v* = b 1 ( c )v*cos(v* c ) +(a 1 ( c )v* 2 + a 3 ( c ))sin(v* c ) (14b) Squaring both equations adding them together, we get (14a) f() = 3 + c 1 ( c 2 + c 2 ( c c 3 ( c (15) where = v* 2 c 1 ( c a o ( c 2 - a 1 ( c 2-2b o ( c 2, (16a) c 2 ( c b o ( c 2-2a o ( c a 2 ( c - b 1 ( c 2-2a 1 ( c a 3 ( c (16b) c 3 ( c a 2 ( c 2 - a 3 ( c 2 (16c) It should be noted that the coefficients c 1 ( c, c 2 ( c c 3 ( c are real. The value of the critical point c is usually determined from the requirement that u( c = 0. In the method used here, the critical point is determined from the requirement that at least one root of Eqn. (15) be real positive, otherwise v = o ( o being the root of the equation) would be imaginary. The existence of an imaginary part of the eigenvalues depends on whether equation has a positive real root. We now state the conditions under which Eqn. (15) has a positive real root. LEMMA. Let be the two turning points of f(), i.e., the roots of df()/d = 3 2 + 2c 1 () + c 2 () = 0 = f()f(). i. If c 3 () < 0, then f() has at least one positive real root. ii. If c 3 () 0, c 2 () < 0 < 0, then f() has positive simple roots. iii. If c 3 () 0, the necessary conditions for f() to have no positive real roots are either a. c 1 () 2 < 3 c 2 () b. c 1 () 2 = 3 c 2 () c. c 1 () 2-3 c 2 () > 0 > 0 or d. c 1 () 2-3 c 2 () > 0 < 0, c 1 () > 0 c 2 () > 0. Proof of the above Lemma is given in the Appendix I. Finally, we have to show that du(/d. This is done by taking the total derivative of Eqns. (12) (13) with respect to then setting = c. Doing this, we get du( = c) dv( = c) B + C + D = 0 (17) d d du( = c) dv( = c) -C + B + E = 0 (18) d d with B = (2a 1 v*- c b 1 v*)sin(v* c *) + (b 1 +*(a 1 v* 2 +a 3 ))cos(v* c *) - 3v* 2 + b o, (19a) C = (b 1 + c (a 1 v* 2 +a 3 ))sin(v* c ) (2a 1 v*- c b 1 v*)cos(v* c ) 2a o v*, (19b) D = ((a 1 v* 2 +a 3 )v*+b 1 v*)sin(v* c ) + (b 1 v*- a 1 v* 2 a 3 )cos(v* c ) + a 2 a o v* 2 (19c) E = (a 1 v* 2 +a 3 -b 1 v* 2 )sin(v* c ) + ((a 1 v* 2 +a 3 )v*+b 1 v*)cos(v* c ) + b o v* (19d) where the prime denotes a derivative with respect of a o, a 1, etc which are then evaluated at c. Applying Crammer s rule to Eqns. (17) (18), we get where du( c ) EC BD d 2 2 B C (20) EC BD = v* 2 {3v* 4 + 2(a o 2 a 1 2 2b o )v* 2 + (b o 2 b 1 2 2(a o a 2 + a 1 a 3 ))}. (21) For du( c /d, EC BD = v* 2 g(v*) 0 so g(v*) = 3v* 4 + 2(a 2 o a 2 1 2b o )v* 2 + (b 2 o b 2 1 2(a o a 2 + a 1 a 3 )) 0. (22) Looking at the Lemma, we see that g(v*) is equal to df()/d, evaluated at = c. The roots of df()/d = g(=v*) = 0 are the two turning points, with = c. Since one of the requirements we have established for the existence of imaginary part of the eigenvalue is f()f() < 0, v* can not be either ( c ) or ( c ), otherwise the product would be equal to International Scholarly Scientific Research & Innovation 1(10) 2007 493

World Academy of Science, Engineering Technology International Journal of Mathematical Computational Sciences International Science Index, Mathematical Computational Sciences waset.org/publication/12037 zero not less than zero. Thus, EC - BD 0 so for the model described by Eqns. (3a) (3c), we have du( c /d provided a critical point exists in the interval of allowed. All of the above discussion can be collected together to form a new theorem; Theorem 3. Concerning the endemic state of our system, Eqns. (3a) (3b), if either i. c 3 () < 0 or ii. c 2 () < 0 () <0, for all [0, lnr o )/ v ), then a Hopf bifurcation can arise as passes through c where c is the critical time delay that satisfies conditions i, ii, iii of Theorem 2. Theorem 3 differs from Theorem 2 by the additional requirement that [0, (lnr o )/ v ). This condition insures that the endemic state is an allowed state, one where all the population densities are positive. The fact that c 2, c 3 are functions of leads to some new consequences. To see why, we note that it is possible to find a value of () in the interval [ 0, (lnr o )/ v ) for which either c 2 (*) = 0 or c 3 (*) = 0 or (*) = 0. One of these * will divide the interval into two sub intervals, [0, *) [, (lnr o )/ v ) in which different conditions can hold. We can then ask, what are the conditions that we can impose on the two regions which will affect the stability or instability of the endemic state? This question has been touched on by Xiao Chen [5]. The answer is given by the following theorem. Theorem 4. Suppose the interval [0, (lnr o )/ v ) is divided into two sub intervals by, [0, *) [, (lnr o )/ v ). If in the interval [0, *), only conditions i. or ii. (at least one positive real root of Eqn. (15) exist) of the Lemma is satisfied in the interval [, (lnr o )/ v ), only condition iii. (there is no positive real root) is satisfied, then there is no critical value of in the interval [0, (lnr o )/ v ) at which a bifurcation occurs. The proof of this theorem is given in Appendix II. This theorem allows us to determine if a bifurcation is possible for a given set of parameter values. IV. NUMERICAL RESULTS AND CONCLUSIONS IVa Realistic Parameter Values. Our bifurcation analysis begins with picking the values of the parameters in our model. The endemic state will be a stable spiral node if the basic reproduction number R > 1 (defined by Eqn. (6)). Its actual value can be determined from observations. If T 2 is the observed doubling time during the initial stage of the epidemic, then R = {(ln T 2 )/ v +1}. Based on the measured doubling times in the growth of infected people during the 1990-91 dengue fever endemic in Sao Paulo State, Marques et al., [10] determined the basic reproduction numbers for twelve cities in the state to be between 1.6 2.5. The values of the parameters picked should be such that if we substitute the values into the expression for the basic reproduction number, Eqn. (6), we should obtain a value of R of the same magnitude as the values observed in nature, i.e., in the range 1 < R < 10. The values of some of the parameters in the model are dictated by reality, e.g. the death rates of the humans mosquitoes, the duration of the infectious period in the human, etc. As we have pointed out, a person infected with the dengue virus is only infectious during the viremia period, which lasts around three days. The recovery rate should be equal to 1/3 per day not the inverse of the length of the illness. The values of the parameters determined by nature are h = 0.000039 per day, corresponding to a life expectancy of 70 years; v = 0.059 per day, corresponding to a mosquito mean life of 17 days b = 1. While one full bite provides enough blood meal for three days, the eating habbits of the Aedes aegypti Ae. albopictus mosquitoes are such that the meal can be interrupted by the slightest movement of the blood provider. Therefore, it takes more than one bite per three days to get a full meal. We have assumed it takes three bites to get a full meal, giving b = 1. The values of the other parameters must be such their substitution into eqn. (5) yields a R in the desired range. Since we will be treating as the bifurcation parameter, we first look at the case of = 0. The basic reproduction number would now be given by eqn. (5d), R o. Using the following values of h, v m, 0.5, 0.75 0.1, respectively, we get a R = 1.91. We have numerically solved Eqns. (3a) to (3c) using the values of the parameters given. In Figure 1, we have plotted the trajectory of the solution in the I v -I phase space we indeed see a stable spiral node. We will now determine whether the system can undergo a Hopf bifurcation to a limit cycle as is increased. As we just showed, the endemic state is stable when = 0. Theorem 4 will be used to establish whether a critical value c (the point at which the stable endemic state loses its stability trajectory becomes a limit cycle) exists in the interval [0, 11.02). The number 11.02 is just the value of (lnr o )/ v. In Figure 2, we have plotted the values of c 2 (), c 3 () () as a function of, using the same numerical values for the other parameters. As we see from Figure 2, c 3 () 0 for [0, 11.02), c 2 () < 0 for [0, 11.014) 0 for [0, 4.48) [11.018, 11.02). Some of the conclusions drawn from Figure 2a. could also be obtained analytically. If we evaluate c 2 (), c 3 () () at = (lnr o )/ v, we find c 2 ( = (lnr o )/ v ) = h 2 ( h + v + r) 2 > 0, c 3 ( = (lnr o )/ v ) = 0 4 2 4 2 lnr (2 r) ( r) ( r) o h h v h v v 0 27 v These are the same results we obtained by looking at Figure 2. International Scholarly Scientific Research & Innovation 1(10) 2007 494

World Academy of Science, Engineering Technology International Journal of Mathematical Computational Sciences Looking at Figure 2., we can identify as being 4.48. This will given us two sub intervals I 1 = [0.4.48) I 2 = [4.48, 11.02). In I 1, c 2 < 0 < 0 while in I 2, condition iii. of the [11.018, 11.02), < 0 c 2 > 0]. Therefore by Theorem 4, there is no critical delay time in the analysis of the underlying International Science Index, Mathematical Computational Sciences waset.org/publication/12037 1.a 1.b 1.c Fig. 1. Stable Spiral Trajectory. Numeric-ally solving Eqns. (3a) - (3b) for = 0 using the following set of parameter values { h = 0.5, v = 0.75, h = 0.000039, v = 0.059, r = 0.33, b = 1 m = 0.1}. With this set of values, R o = 1.91 which is within the range of biological acceptable values. As we see, the trajectories in the I v -S I-S phase plane spirals into the endemic equilibrium state. The trajectory in the I v -I also spirals in, although this is not clearly evident. Lemma holds, i.e., no positive real root of eqn. (15) exist [Note that for [4.48, 11.018), > 0 c 2 < 0 for interval [0, (lnr o )/ v ) when the above values of the parameters are used. The above conclusion is not surprising. It has been pointed out that the limit cycles predicted by a mathematical 2.a 2.b 2.c Fig. 2. Dependence of c 2 (), c 3 () () on the time delay using realistic values of the parameters. The values used are the same as those used to generate Figure 1, i.e., { h = 0.5, v = 0.75, h = 0.000039, v = 0.059, r = 0.33, b = 1 m = 0.1}. The incubation period is varied between 0 11.02 which is the value of (ln 1.91)/ v. As is seen, c 2 () is negative for < 11.01, c 3 () is positive over the entire range of () is negative in the range [0, 4.48), positive in [4.48, 11. 018) negative again in [11.018, 11.02). The point * divides the entire interval into two sub intervals, I 1 I 2. In I 1, the conditions for a positive real root of eqn. (12) while in I 2, the conditions for eqn. (12) to have no positive real roots hold. equation does not always occur when biologically relevant values are used. In a seminal paper, Caughley [11] proposed a simple predator-prey model to describe an elephant-tree ecosystem. He found that the elephants trees could coexist in a stable limit cycle with a frequency of oscillation of approximately 270 years. Duffy et al., [12] has questioned this hypothesis. They found that using realistic parameter values, limit cycles were highly unlikely. Instead, their International Scholarly Scientific Research & Innovation 1(10) 2007 495

World Academy of Science, Engineering Technology International Journal of Mathematical Computational Sciences International Science Index, Mathematical Computational Sciences waset.org/publication/12037 parameter analysis showed that an equilibrium state was more likely. In their paper, Khan Greenhalgh [4] found that while a Hopf bifurcation was theoretically possible in their model, it would not occur under most realistic conditions. For a bifurcation to occur, the disease that they would be vaccinating against would be one for which the chances of dying from the diseases are greater than that of recovering. In addition, the fraction of newborns, who are effectively vaccinated at birth (immunity being passed from the mother) is just below its critical threshold value. Such a disease probably does not exist. IV.b Unrealistic Parameter Values. To determine whether it is possible that there are parameter values such that a Hopf bifurcation is possible, we have picked a set of parameter values ( h,, v, h, v,b,r,m) for which the basic reproduction number would be unphysical, i.e., not between 1 3. The set chosen (0.00001, 1.0, 0.0000456, 0.058, 170,0.02, 0.2 gives a R o = 57.67. We have plotted on Figure 3, the values of c 3 () over the range 0 < < 81.09, the value of (lnr o )/ v. While we have varied up to 81.09, should have been varied up to 17, the life time of the mosquito. Fig. 3. Dependence of c 3 () on the time delay using parameter values which produce an unrealistic R o. The values of the parameters used are h = 0.00005, v = 1, h = 0.000046, v = 0.05, r = 0.02, b = 170 m = 0.2. These values give a R o = 57.67. We have varied between 0 81.09 (ln 57.67)/ v ). As we see, c 3 () < 0 over the entire range of..if the mosquito dies before it becomes infectious, the transmission of the disease can not be maintained the epidemic ceases. As we see, c 3 () is negative over the entire range of. According to Theorem 3, the critical delay c exist in the interval {0, 81.09]. We then systematically varied (keeping the same values for the other parameters) starting from = 0 attempted to solve eqn.(15). Doing this, we found *= 9.5. To truly see whether the system bifurcates, we have solved Eqns. (3a) - (3b) for = 9.0 < *. The trajectory of the solution in the I v -I phase space is plotted in Figure 4a. As we see, the solution spirals into the equilibrium point (0.809, 0.142, 0.002). Changing to the critical value 9.5, we get a stable limit cycle orbit (Figure 4b). Increasing to beyond the critical value, i.e., = 10 > *, we get the trajectory shown in Figure 4c. It is an outward spiral going away from an unstable equilibrium point. As we have mentioned, the endemic states depend on the time delay. At = 9.5, the equilibrium state is (0.805, 0.143, 0.002). It becomes an unstable point located at (0.801, 0.144, 0.002) when = 10. Fig. 4. Trajectories of (I v (t), I(t)) for three values of. Numerical solutions of Eqns. (3a) (3b) using the parameter values used to obtain Figure 3. The critical value of was found to be * = 9.5. (4a) Stable spiral trajectory when = 9.0 < * is used. (4b) Limit cycle orbital is obtained when is set to (9.5). (4c) Unstable spiral trajectory when = 10.0 > *. The two most obvious unrealistic parameters are the biting rates of the mosquitoes the recovery rate of the human. For the system to bifurcate, we need a mosquito that would bite 170 times a day. Even though the Ae. Aegypti mosquito is a mosquito which will take a new bite if her meal is interrupted, 170 interruptions would be too much to expect. A recovery rate of 0.02 per day requires the viremia stage last for at least fifty days. This is not a characteristic of dengue infection. In addition, the choice of parameters used is unrealistic since they would also lead to an unrealistically large number of secondary infections. International Scholarly Scientific Research & Innovation 1(10) 2007 496

World Academy of Science, Engineering Technology International Journal of Mathematical Computational Sciences International Science Index, Mathematical Computational Sciences waset.org/publication/12037 IV. DISCUSSION The annual cycle seen in the incidence of dengue fever (dengue Hemorrhagic fever) in Bangkok, Thail between 1966 1998 by Hay et al.,[13] is not indicative of a limit cycle. We have shown that the appearance of a limit cycle in the transmission cycle of dengue fever is highly unlikely. The annual cycles arise from the seasonal variations, which occur in many of the parameters in the model. Dowell [14] has classified the causes of these cycles into three groups: pathogen appearance disappearance, environmental changes host behavior changes. Statistical significant correlation s between epidemic cycles cycles of temperature, humility, rains or winds have been found. Dowell has pointed out that the seasonal variations should be distinguished from the periodic behaviors, which would be intrinsic to the model. Hay et al., also drew attention to this when they remarked that the focus of future research on mosquito borne diseases should be on combining the extrinsic (climate changes) determinants with the intrinsic determinants. ACKNOWLEDGEMENTS One of the authors (IMT) would like to thank the Thail Research Fund (TRF) for financial support. REFERENCES [1] D.J. Gubler DJ, Dengue Dengue Hemorrhagic Fever, Clin. Mirobiol. Rev. 11 (1998) 480. [2] World Health Organization, Dengue hemorrhagic fever: diagnosis, treatment, prevention control, 2 nd Ed. (1997) WHO.. [3] A. Martin S. Ruan, Predator-prey models with delay prey harvesting. J. Math. Biol. 43 (2001) 247. [4] Q.L.A. Khan, D. Greenhalgh, Hopf bifurcation in epidemic models with a time delay in vaccination. IMA J. Math. Appl. Med. Bio. 16 (1998) 113. [5] Y. Xiao, L. Chen, Modeling analysis of a predator-prey model with disease in prey. Math. Biosci. 171 (2001) 59. [6] S. Ruan, J. Wei, On the zeros of a third degree exponential polynomial with application to a delay model for control of testosterone secretion. IMA J. Math. Appl. Med. Biol.18 (2001) 41. [7] J. Tam, Delay effect in a model for virus replication. IMA J. Math. Appl. Med. Biol. 16 (1999) 29. [8] J.E. Marsden, M. McCracken, The Hopf Bifurcation Its Application, (Springer-Verlag, Berlin(1976)). [9] L. Esteva, C. Vargas C, Analysis of a dengue disease transmission model. Math. BioSci. 150 (1998) 131. [10] C.A. Marques, O.P. Forattini, E. Massad,. The basic reproduction number for dengue fever in Sao Paulo state, Brazil: 1990-1991 epidemic. Trans. Roy. Soc. Trop. Med Hyg. 88 (1994), 58. [11] G Caughley, The elephant problem-an alternative hypothesis. East Aft. Wildl. J. 14 (1976) 265. [12] K.J. Duffy, B.R. Page, J.H. Swart, V.B. Bajic, Realistic parameter assessment for a well known elephant-tree ecosystem model reveals that limit cycles are unlikely. Ecol. Mod. 121 (1999) 115. [13] SI Hay, MF Myers, DS Burke, DW Vaughn, T Endy, N Ana, GD Shanks, RW Snow, DJ Rogers, Etiology of interepidemic periods of mosquito-born disease. PNAS 97 (2000) 9335. [14] S. Dowell, Seasonal variation in host susceptibility cycles of certain infectious diseases. Emer. Inf. Dis. 7 (2001) 369. [15] Y. Kuang, in: Delay differential equations with appli-cation to population dynamics, (Academic Press, New York, 1993) page 66. APPENDIX I Poof of the Lemma: To establish condition i., we first note that with c 3 () < 0, f(=0) would be negative. At, f() would be positive. F() would have to cross the f() = 0 axis in the interval [0,], thus showing that at least one of the roots is real positive. Condition ii. is more complicated to established. If c 2 () < 0, one of the turning points (say ) would be positive while the other (say ) would be negative. The condition < 0 means that the sgn f() = - sgn f(), i.e., if f() is in the upper half plane, then f() is in the lower half plane or vice versa. The condition c 3 () > 0, means that f(0) would now be positive. f() would remain positive f(-) would be negative. Since is the positive valued turning point, the signs of f(0) f() requires that f() be in the lower half f() plane. The f() would have to cross the f() = 0 axis at two points somewhere between 0 between. Since the coefficients c 1 (, c 2 ( c 3 ( are real numbers, the roots of f() = 0, the roots must be either all real or one real a pair of complex conjugates. If z 1, z 2 z 3 are the three roots, then c 3 ( = - z 1 z 2 z 3. If c 1 () 2 < 3 c 2 (), then there is only one real root. Calling this root z 1, we get z 1 {(Re z 2 ) 2 +(Im z 2 ) 2 }. Since this is just c 3 () c 3 () 0, z 1 must be negative. The condition c 1 () 2 = 3 c 2 () leads to two of the roots being double roots leaving c 3 () = -z 1 z 2 2. For this to be greater or equal to zero, z 1 must again be negative. The conditions c 1 () 2 3 c 2 () > 0 > 0 means that two turning points exist that both f() f() lie in the same (upper or lower half) plane. Now, however, f() does not cross the f() = 0 axis between, meaning that we do not have a positive real root. The last condition for the non existence of a positive real root, c 1 () 2 3 c 2 () > 0 < 0, c 1 () > 0 c 2 () > 0 is established by noting that the two turning points appear in the left half plane. The crossing of f() with the zero axis occurs between 0 - at a, leading the root to be negative. APPENDIX II Proof of Theorem 4. We first introduce three delay times, i *, j * k * defined by c 2 ( i *) = 0, c 3 ( k *) = 0 ( j *) = 0. There may be more than one value of each of these critical delays since c 2 ( i *) = 0, c 3 ( k *) = 0 or ( j *) = 0 could have more than one root. * could be either i *, j * or k *. Picking it to be max{ i *, j *, k *} or to be min{ i *, j *, k *}, we find that it can not be in the region satisfying any of the conditions of the Lemma (i.e. all the conditions are expressed as being less than, not equal to zero). I 1 I 2 are the intervals [0, *) [*, (lnr o )/ v ), respectively. The roots of Eqn. (8) can be viewed as a continuous function of the time delay, i.e., [See Lemma 5.1 in ref. 5 or Theorem 1.4 in the textbook of Yuang Kuang [15]]. At = 0, we know that the endemic state is asymptotically stable (by Theorem 1). Therefore the International Scholarly Scientific Research & Innovation 1(10) 2007 497

World Academy of Science, Engineering Technology International Journal of Mathematical Computational Sciences International Science Index, Mathematical Computational Sciences waset.org/publication/12037 Re{(=0)} would be negative = 0 would lie on the left h side (LHS) of the complex -plane. As moves from I 1 to I 2, would enter into the right half plane (RHS) by crossing the imaginary axis. Then the endemic state would be unstable Re{( > *)} would be positive. Now consider what will happen if the critical time delay c lies in I 1. Since the transversality condition holds in I 1, a bifurcation will occur at c. Therefore, as passes c, the endemic state becomes unstable, i.e., Re{ c I 1 } would be in the RHS of the complex -plane. As approaches *, ) should be approaching the boundary from the right d ( c ) 0, d it cannot cross the imaginary axis to get into the RHS since it is already in the RHS. This inconsistency requires that there be no c in I 1. If, however, there were two critical time delays in I 1, the present arguments would not apply. Our theorem only applies if there is only one (or an odd number) critical point. To complete the proof, we consider what would happen if c I 2, notwithsting the fact that the definition of I 2 precludes the presence of c in the interval. But first we look at the stability for close to (lnr o )/ v. For I 2, is on the RHS of the complex -plane. As (lnr o )/ v, the basic reproduction number Roe v 1 Also in this limit, S* 1, I*0 I v * 0. The characteristic equiation, Eqn. (8) reduces to ( 2 + a o + b o ) = 0 where a o = 2 h + v + r b o = h ( h + v + r). It is easily shown that the two non-zero roots of this equation are real negative. This means that as (lnr o )/ v, () must head towards the imaginary axis from the right. We now place c into I 2. As passes c, () should cross the imaginary axis move into the LHS of the complex -plane. As we have stated before, as (lnr o )/ v, () should be moving to the left to get close to the imaginary axis. With c present in I 2, ( > c ) is already on the LHS there is no need to move towards the imaginary axis. This contradicts the original requirement. To resolve this, we need c I 2. Since c I 1 also, c I 1 I 2 no bifurcation is possible in the interval 0, (lnr o )/ v ) if the interval can be divided into two sub intervals, one in which conditions i. ii. of the Lemma hold the other in which condition iii. holds. (Again we point out that the theorem proof do not apply if there are two critical time delays in the interval I 1. ) International Scholarly Scientific Research & Innovation 1(10) 2007 498